Direct Democracy and Political Equality in the American States

Transcription

1 Direct Democracy and Political Equality in the American States Patrick Flavin Assistant Professor Department of Political Science Baylor University Prepared for presentation at the annual meeting of the Western Political Science Association in Hollywood, CA, March 28-30, A previous version of this paper was prepared for presentation at the 2012 meeting of the American Political Science Association in New Orleans, LA (conference cancelled due to weather).

2 Abstract Whose opinions are reflected in the policy decisions made by state governments? A growing literature in political science documents unequal political representation at the national level, but political inequality at the state level remains under examined. This omission is unfortunate because the rich variation in political behavior, laws, institutions, and culture across the fifty states provides unique leverage for investigating what conditions lead to more or less political equality. Using public opinion measures from the National Annenberg Election Surveys and data on state policy outputs, I find that state policy is consistently more proximate to the opinions of citizens with higher incomes. Using this measure of opinion-policy proximity, I generate an index of the equality of political representation (based on citizens incomes) that is comparable across the states. I then evaluate the relationship between various measures of direct democracy and political equality and find evidence that states where it is easier to place a measure on the ballot for popular vote and states where the ballot initiative is frequently used tend to weigh citizens opinions more equally in the policymaking process. These findings underscore the importance of laws and institutional design in promoting political equality in the United States. Keywords: Political inequality, political representation, direct democracy, public opinion, policy

3 Government responsiveness to citizens political opinions is central to democratic theory and the idea that government ought to be, as Abraham Lincoln remarked, of the people, by the people, for the people, is a guiding principle for American democracy. This belief in popular sovereignty that elected officials should be directly accountable to the people for their policy decisions tends also to be the way Americans judge the health and legitimacy of other democracies around the world. Political representation is, simply put, a yardstick by which the quality of a democracy can be measured. Toward this end, a large literature in political science has examined the congruence between citizens aggregated political opinions and their elected officials policy decisions in the United States (e.g., Page and Shapiro 1983; Erikson, Wright, and McIver 1993; Monroe 1998; Erikson, MacKuen, and Stimson 2002; Lax and Phillips 2012; for a review see Shapiro 2011). However, any evaluation of the quality of a democracy must also be concerned with e quality. As Robert Dahl (2006, ix) states, The existence of political equality is a fundamental premise of democracy. In the context of political representation, political equality refers to the extent to which citizens have an equal voice in governmental decisions. One of the bedrock principles in a democracy is the equal consideration of the preferences and interests of all citizens (Verba 2003, 663). In short, political equality the equal weighting of citizens opinions when elected officials make important policy decisions is also central to any assessment of democratic performance. 1 Although political equality has remained a paramount normative concern for decades, political scientists have devoted little attention to empirically examining unequal political 1 Verba and Orren (1985, 8) add that democracy implies a certain degree of political equality if not full equality of political representation among citizens, at least some limit to political inequality. 1

4 representation in the United States. As the American Political Science Association Taskforce on Inequality and American Democracy (Jacobs and Skocpol 2005, 124) lamented: Unfortunately, political scientists have done surprisingly little to investigate the extent of actual inequalities of government responsiveness to public opinion that is, whether distinct segments of the country exert more influence than others. However, a series of recent studies have responded to this challenge and found that affluent citizens exert more political influence than disadvantaged citizens in the formulation of public policies (Gilens 2005, 2012; Jacobs and Page 2005; Bartels 2008; Rigby and Wright 2011; Ellis 2012; Flavin 2012). The existence of unequal political representation presents scholars a question with very concrete policy implications: What sort of laws and institutional arrangements can help promote a more equal weighting of citizens opinions in the policymaking process? For example, do certain campaign finance reforms like stricter disclosure or contribution limits laws lead to a more equal weighting of citizens opinions? What about the use of voting and registration reforms like early and absentee voting or Election Day registration? By studying how certain institutional arrangements lead to more or less political equality, scholars can shed light on an important normative question for American democracy. Unfortunately, most previous studies of unequal democracy examine political influence at the national level (but see Gilens, Lax, and Phillips 2011; Rigby and Wright 2011; Flavin 2012), where little institutional variation exists. In contrast, the American states provide rich variation in opinions, policies, institutions, and conditions (Jewell 1982). As Erikson, Wright, and McIver (1993, 2) argued in their landmark study of political representation in the states, Almost all U.S. studies of the influence of public opinion focus on the national level. However, the ideal place to investigate the relationship between public opinion and public policy would 2

5 seem to be the American states. With fifty separate state publics and fifty sets of state policies, the states provide an ideal laboratory for comparative research. In this paper, I use the variation across the American states to study the impact of direct democracy laws on the equality of political representation for the rich and poor. I first find that states with the ballot initiative process are no more or less politically equal compared to states without the initiative process. However, when the analysis is confined only to states with the initiative, state policies are more equally representative of all citizens opinions in states where it is easier to place a measure on the ballot for popular vote and states where the initiative process is heavily used. These results suggest that having and frequently using the initiative process may be a viable avenue for ensuring that the opinions of disadvantaged citizens are represented in the political arena. More broadly, these findings underscore the importance of laws and institutional design in promoting political equality in the United States. Background Political scientists and political observers more generally have long warned that political representation in the United States is tainted by an upper class bias such that wealthier citizens have more influence over government policy decisions than the poor (e.g., Schattschneider 1960; Dahl 1961). While political scientists have devoted considerable attention to documenting unequal political participation, or inputs into the political system (Piven and Cloward 1988; Hill and Leighley 1994; Verba, Schlozman, and Brady 1995; Verba 2003), most have tended to shy away from actually assessing whether elected officials equally weigh their constituents 3

6 opinions when making important policy decisions. 2 As Larry Bartels (2008, 253) aptly points out, For the most part, scholars of political participation have treated actual patterns of government responsiveness as someone else s problem. A series of recent studies have sought to correct this problem and more fully understand unequal political representation in the United States. Jacobs and Page (2005) use parallel opinion surveys of the public and political elites to show that internationally oriented business leaders leverage more influence over American foreign policy decisions than the opinions of the general public. Gilens (2005, 2012) collects data from individual public opinion poll questions across a wide array of political issues and finds that subsequent federal government policy decisions disproportionately reflect the views of the affluent, and this is especially true when the preferences of the rich and poor diverge. He concludes that congruence between the political opinions of the poor and government policy tends to arise only in instances where the poor share similar attitudes with the wealthy. Bartels (2008) examines the link between political factors and growing economic inequality and demonstrates that the opinions of affluent constituents strongly predict the voting behavior of their Senators (both their revealed general voting ideology and specific roll call votes) while the opinions of those with low incomes display little or no relationship. Ellis (2012) reports that these same findings also extend to the House of 2 One reason for the large literature on unequal levels of political participation is that it is relatively easy to measure in a way that inequalities in political representation are not. As Verba and Orren (1985, 15) point out: Political equality cannot be gauged in the same way as economic inequality. There is no metric such as money, no statistic such as the Gini index, and no body of data comparing countries. There are, however, relevant data on political participation. 4

7 Representatives. In short, this emerging literature points to unequal democracy in American politics. 3 Amidst growing evidence at the national level, scholars are only beginning to identify and investigate unequal political representation at the state level. 4 Rigby and Wright (2011) uncover evidence that the general ideological tone of state economic policies tends to be most responsive to the opinions of the rich and hardly at all to the poor, and that this is particularly true in poorer states. 5 Flavin (2012) finds that citizens with low incomes tend to have little influence on state policy incomes measured both as general policy liberalism and specific social policies like the death penalty, abortion, and gun control. Gilens, Lax, and Phillips (2011) also find some evidence that citizens with low incomes are underrepresented in the state 3 However, a set of recent studies have called these findings of unequal political representation into question (Soroka and Wlezien 2008; Ura and Ellis 2008; Erikson and Bhatti 2011). These studies argue that because so little variation in political preferences exists across income groups, it is nearly impossible to distinguish whose opinions are being reflected in the public policy decisions made by elected officials. 4 At the city government level, an earlier study of public opinion and policy in 51 American cities found that city policies tended to respond most to the opinions of citizens with higher socioeconomic status (Schumaker and Getter 1977). In contrast, Berry, Portney, and Thomson (1993) found little economic bias in policy responsiveness for the cities they studied. 5 Rigby and Wright (2011) derive separate summary measures of citizens general economic and social attitudes by (after imputing a significant amount of missing data across survey items) factor analyzing multiple opinion items from the 2000 and 2004 National Annenberg Election Surveys and collapsing the mean opinion measure of low, middle, and high income respondents within each state. They then derive separate summary measures of state policy on economic and social issues by factor analyzing a set of state policies for each area. 5

8 policymaking process compared to the more affluent. In sum, a growing body of literature has documented that the unequal political representation found at the national level extends to the states as well. Despite this growing literature, our understanding of the precise mechanisms that lead to more or less political equality remain limited. As discussed above, studies of unequal political representation at the national level run into the methodological roadblock that there is little institutional variation that might help explain why some citizens exert more political influence than others. In contrast, the fifty states provide rich variation in terms of laws, institutions, political culture, and a host of other factors. Moreover, the information learned from studies that use the states as comparative units of analysis can help to inform policymakers interested in institutional reforms that might promote greater political equality. This paper uses the variation in direct democracy laws across the states to examine the effect of the ballot initiative on the equality of political representation. Besides indirectly influencing public policy through the election and monitoring of public officials, citizens in some states can exert direct influence on policy decisions through the ballot initiative process. Currently, twenty-four states allow citizens to present a petition to place a measure on the ballot for popular vote at election time, generally with the requirement that a certain number of signatures accompany the petition. 6 Table 1 lists the states that allow the initiative, the year they adopted the process, and the signature requirements for placing a measure on the ballot. Looking down the list, two factors jump out: most states (1) adopted the initiative process during the Progressive Era (1900s and 1910s) and (2) require signatures from roughly 5-10% of the 6 Three states only allow citizens to propose constitutional amendments, six states only allow citizens to propose regular statutes, and fifteen states allow citizens to propose both. 6

9 voting population to place a measure on the ballot for popular approval. In recent years, citizens in initiative states have had the opportunity to directly decide state policy on a variety of issues including same-sex marriage laws, collective bargaining rights for public employees, the permissibility of medical marijuana, and income and property tax rates. [Table 1 about here] Proponents of the initiative process argue that it shifts the power in the policymaking process away from political elites and towards ordinary voters. As Matsusaka (2004, 71) writes: Without the initiative, voters are forced to accept the policy choices of the legislature. With the initiative, voters are given choices. Recent empirical studies have generally confirmed this belief that direct democracy allows citizens greater influence over policy by documenting that the presence of the initiative process strengthens the congruence between aggregated public opinion and state public policies (Arceneaux 2002; Matsusaka 2004, 2010; Burden 2005; but see Lascher, Hagen, and Rochlin 1996; Camobreco 1998). Moreover, even the threat of citizens using the initiative process is often enough to induce responsiveness from potentially out of step state legislators (Gerber 1996, 1999). In short, direct democracy seems to improve policy responsiveness to aggregated public opinion in the states. In this paper, I extend this line of inquiry and ask: Does direct democracy enhance the equality of political representation? Theoretically, the initiative process should allow citizens with limited means a greater voice in the political process because they get a direct say on proposed policy change and all citizens, no matter how rich or poor, only get to cast a single vote. This is in contrast to the regular policymaking process in the state legislature, where the opinions of wealthier citizens likely exert greater influence on legislators because of the time and 7

10 resources usually required to successfully lobby government officials (Hall and Wayman 1990; Evans 1996). On the other hand, the initiative process requires rather intense citizen involvement, especially when attempting to collect enough signatures to get a proposal on the ballot. This intensive political activity is likely organized by those with more ample political resources (i.e. those with higher socioeconomic status). If citizens who participate in politics at higher levels are the driving force behind the initiative process, propositions that actually make it onto the ballot may tend to over-represent the opinions and interests of citizens with higher incomes. Because the initiative process can be expensive and time intensive to navigate, it may provide even greater political influence for wealthy groups and individuals (Ellis 2002). For example, Broder (2000) argues that direct democracy subverts the will of the people by allowing a powerful few citizens to propose policy changes (also see Smith 1998; Schrag 2004). Additionally, Moore and Ravishankar (2012) report empirical evidence that racial minorities (who, on average, are of lower socioeconomic status compared to whites) in California tend to lose more often than whites when voting on ballot propositions, and that this disadvantage is not limited to a small subset of racially-targeted propositions. So, instead of politically empowering citizens at the bottom of the income distribution, the initiative process may instead be yet another avenue whereby economic inequality is reproduced as political inequality. To date, no study has directly evaluated the relationship between direct democracy laws and the equality of political representation. This shortcoming in our understanding is unfortunate given that direct democracy could potentially be an important mechanism for ensuring that more citizens opinions are represented in the political arena. To further our understanding of the effects of laws and institutional design, I investigate whether the presence and usage of the ballot 8

11 initiative process (measured in various ways) in the states leads to more or less equal weighting of all citizens political preferences. Evaluating the Equality of Political Representation in the American States Although political representation is central for American democracy, there is little consensus on how best to measure the concept. For years, political scientists have experimented with different methods of assessing the link between the people and their government (Miller 1964; Achen 1978; Wright 1978; Powell 1982; Burden 2004; Gershtenson and Plane 2007; Griffin and Flavin 2007). One important distinction (Weissberg 1978) has been whether public opinion is compared to the behavior of individual elected officials (i.e. dyadic representation ) (Miller and Stokes 1963; Achen 1978; Powell 1982; Bartels 1991; Clinton 2006) or to government policy decisions more broadly (i.e. collective representation ) (Page and Shapiro 1983; Erikson, Wright, and McIver 1993; Erikson, Mackuen, and Stimson 2002; Wlezien 2004). I focus on the latter because government policy is the final link of the chain that begins with citizens inputs (their political opinions and behaviors) into the political system. More importantly, regardless of how a citizen s particular state house member or state senator votes on any given bill in the state legislature, citizens are ultimately affected by the decisions of the legislature as a whole and the actual policies that are implemented. Policy representation is measured using a proximity technique that places public opinion and policy on the same linear scale and compares the distance between the two (Achen 1978). Using this method, as the ideological distance between a citizen s opinion and policy grows (i.e. policy is ideologically further from a citizen s preferences), that citizen is not well represented. The identical measurement technique has been used in several recent studies to evaluate the 9

12 ideological distance between citizens and Member of Congress (Griffin and Flavin 2007; Griffin and Newman 2007, 2008; Ellis 2012), Senators (Gershtenson and Plane 2007), and presidential candidates (Burden 2004; Jessee 2009) in the United States as well as the ideological distance between citizens and political parties in Europe (Blais and Bodet 2006; Powell 2009; Golder and Stramski 2010; Giger, Rosset, and Bernauer 2012). In practical terms, this proximity technique allows a researcher to evaluate whether a conservative (liberal) citizen lives in a state that, compared to other states, implements conservative (liberal) policies and is well represented, implements liberal (conservative) policies and is poorly represented, or gradations in between. For example, Figure 1 compares two hypothetical citizens, Citizen A and Citizen B, who both live in the same state. When the two citizens political ideologies are placed on the same metric as state policy, there is less ideological distance between Citizen A and state policy as compared to Citizen B and state policy. Under the proximity conceptualization of policy representation, Citizen A is better represented than Citizen B. [Figure 1 about here] To measure ideological proximity, I require two pieces of data: (1) a measure of citizens opinions and (2) a measure of state policy. To measure public opinion, I combine data from the 2000, 2004, and 2008 National Annenberg Election Surveys (NAES), three random digit dialing rolling cross sectional surveys conducted in the months leading up to that year s presidential election. For years, scholars of public opinion in the states have wrestled with the problem of not having enough respondents in public opinion polls to make reliable state-level estimates and inferences. One way to address this problem is to pool surveys over a long period of time and disaggregate the data by state (Erikson, Wright, and McIver 1993). Another way is to simulate state opinion by using national polls and multi-level modeling to derive estimates for the states 10

13 based on demographic characteristics (Park, Gelman, and Bafumi 2006; Lax and Phillips 2009a, 2009b). The major advantage of pooling these three NAES surveys is their sheer sample size which allows a large enough sample without having to aggregate across a long time period or simulate state opinion (Carsey and Harden 2010). This large sample size is especially important because I later assess the relationship between income and ideological proximity within individual states. 7 Citizens general political ideology is measured using the following item from the NAES: Generally speaking, would you describe your political views as very conservative, conservative, moderate, liberal, or very liberal? I code the measure such that it runs from -2 (very conservative) to +2 (very liberal). Data on citizens self-reported political ideology have been commonly used to measure public opinion in previous studies of political representation (e.g., Erikson, Wright, and McIver 1993; Griffin and Flavin 2007; Bartels 2008; Flavin 2012) and there is reason to be confident that self-reported ideology is an accurate measure of citizens aggregated policy-specific opinions. 8 In support of this claim, Table 2 displays the percentage of respondents from the 2000 and 2004 NAES who report a particular opinion categorized by their self-reported political ideology. Looking across the columns, it is clear that respondents who 7 A total of 177,043 NAES respondents across the three survey waves answered the ideological selfplacement and income items. All states except North Dakota (N=475) and Wyoming (N=414) have a sample size of over 500 respondents. Alaska and Hawaii were not surveyed, so all analyses in this paper report results from the remaining 48 states. 8 However, some previous studies have questioned whether a person who identifies him/herself as a liberal (conservative) actually holds liberal (conservative) policy opinions; that is, whether citizens selfreported or symbolic ideology accurately reflects their operational ideology when queried about specific issues (Knight 1985; Jacoby 1995; Jennings 1992; Ellis and Stimson 2009). 11

14 identify themselves as liberal are more likely to report liberal policy opinions. For example, only 38% of respondents who place themselves in the very conservative category believe that Government should reduce income differences between rich and poor. In contrast, fully 77% of respondents who place themselves in the very liberal category support that policy proposal. These differences across ideological classifications suggest that self-reported ideology can be used to approximate citizens underlying policy opinions. [Table 2 about here] Next, to measure public policy, I require a general measure of the liberalism (Klingman and Lammers 1984) of state policy outputs that comports with the survey item that asks citizens their general political ideology. In their seminal book on state opinion and policy, Erikson, Wright, and McIver (1993) developed a composite index of state policy liberalism using eight policy areas for which liberals and conservatives typically disagree. Gray, Lowery, Fellowes, and McAtee (2004) updated this policy liberalism measure for 2000 using the following five policy items: (1) state regulation of firearms as measured by state gun laws; (2) scorecard of state abortion laws in 2000; (3) an index of welfare stringency that accounts for Temporary Assistance to Needy Families (TANF) rules of eligibility and work requirements for ; (4) a dummy measure of state right-to-work laws in 2001; and (5) a measure of tax progressivity calculated as a ratio of the average tax burden of the highest five percent of a state's earners to the average tax burden of the lowest forty percent of a state's earners. 9 These five components are then 9 Gray et al. (2004) argue that using these policy items, as opposed to a measure of per capita expenditures for different policy areas, precludes the possibility that policy liberalism is simply a proxy for a state s wealth. The five measures produce a Cronbach's alpha of

15 standardized and summed in an additive index such that more liberal state policies are coded higher. I use this index as my first measure of the general ideological tone of state policy. Second, a recent article by Sorens, Muedini, and Ruger (2008) provides a rich source of data on state policies in twenty different areas ranging from public assistance spending to gun control to health insurance regulations. 10 In addition to specific statutes and spending data, the authors provide a summary index of policy liberalism for each state that they derive by factor analyzing their entire range of policies. I use this composite score as a second measure of general policy liberalism. 11 Together, the two policy liberalism measures represent the unidimensional liberal/conservative ideology of state policy decisions that correspond well to the measure of citizens general political ideologies described above. Measuring ideological proximity requires a method of placing citizens opinions and state policy on a common scale for comparison (see Figure 1). Drawing on previous studies that have also used a proximity technique to measure political representation (Achen 1978; Burden 2004; Blais and Bodet 2006; Gershtenson and Plane 2007; Griffin and Flavin 2007; Griffin and Newman 2008; Jessee 2009; Powell 2009; Golder and Stramski 2010; Ellis 2012; Giger, Rosset, and Bernauer 2012), this paper approaches this task in three different ways. If all three measurement techniques point to the same conclusion, then we can be more confident in the robustness of the results The state policy data can be accessed online at 11 The Gray et al. (2004) and Sorens, Muedini, and Ruger (2008) policy liberalism measures correlate across the states at One common critique of using the proximity method to evaluate political representation is that, regardless of the statistical technique used to match up the two, opinion and policy are not on the same 13

16 First, I standardize all ideological opinions to a mean of zero and a standard deviation of one and take the two recent measures of general state policy liberalism described above (Gray et al. 2004; Sorens, Muedini, and Ruger 2008) and standardize them as well. After standardizing both opinion and policy, they are now on a common (standardized) metric, similar to the strategy used by Wright (1978) and Ellis (2012). To measure proximity, I take the absolute value of the difference between a respondent s ideology score and the policy liberalism score for his/her state using both of the measures of policy. This creates the first measure of ideological distance for each respondent in the NAES sample which I term the Standardized measure. Second, I rescale the two measures of state policy to the same scale (-2 to +2) as citizens self-reported ideology. This technique is similar to that used in early studies of congressional representation (Miller 1964; Achen 1978) and one that is still advocated by representation scholars today (Burden 2004; Griffin and Newman 2008). I again take the absolute value of the distance between a respondent s ideology score and the policy liberalism score for his/her state and term it the Same Scale measure. Third, I rescale policy to a tighter range (-1 to +1) than citizens ideologies. I do so because we can expect citizens ideological opinions to have a wider range and take on more extreme values compared to actual state policy outputs. This transformation to a tighter scale is suggested and implemented by Powell (1982, 1989) in her studies of congressional scale. However, whatever the flaws of each of the three different measures of ideological proximity in matching up opinion and policy, they are likely equally flawed for all citizens regardless of their income. Therefore, the proximity measures can provide information about, for example, how ideologically proximate opinion and policy are for a poor person relative to a rich person (see Griffin and Newman 2007, Ellis 2012). 14

17 representation. Again, I take the absolute value of the difference between a respondent s ideology score and the state policy liberalism score for his/her state and term it the Restricted Scale measure. Together, there are three different measurement techniques and two different measurements of state policy liberalism, for a total of six different measures of ideological proximity between citizens opinions and state policy. I am then interested in whether there are systematic differences in proximity between opinion and policy across citizens. Specifically, I am interested in whether there is a link between a citizen s income and the ideological distance between opinion and policy. Because I am interested in unequal political representation within each state and state populations can vary widely in terms of their income distribution, it would be unwise to simply compare the incomes of citizens in one state to the incomes of citizens in another state. Simply put, we might expect someone making $100,000 per year living in West Virginia to exert comparatively greater political influence than someone making $100,000 per year living in Connecticut. To account for differences in the income distribution across states, I generate a measure of state relative income that compares a respondent s income with the average income for a resident in his or her state. A positive score for state relative income indicates that a respondent is above the mean while a negative score indicates that a respondent is below the mean. Armed with this new measure, I then assess whether there is a systematic relationship between a citizen s (state relative) income and the ideological distance between their opinion and state policy. To evaluate this relationship, I regress the measure of ideological distance on (state relative) income for every respondent in the sample using the six different measures of 15

18 ideological proximity described above. 13 The results of these six regressions are reported in Table 3. Reading across the six columns reveals strong evidence of unequal political representation. Specifically, all six coefficients are negative and bounded below zero which indicates that as a respondent s (state relative) income increases, the distance between their ideology and state policy decreases and they are better represented. Put another away, the lower a respondent s income, the greater the distance between opinion and policy and the worse that respondent s general political ideology is represented in the general liberalism of his or her state s public policies. Substantively, the larger opinion-policy distance for a respondent at the 10 th percentile for (state relative) income compared to a respondent at the 90 th percentile is about the same as the difference between a respondent at the 10 th percentile for (state relative) level of education compared to the 90 th percentile (Gilens 2005) and larger than the difference between an African American respondent compared to a white respondent (Griffin and Newman 2008). These findings comport with the small but growing set of studies (Gilens, Lax, and Phillips 2011; Rigby and Wright 2011; Flavin 2012) that have found that disadvantaged citizens are systematically underrepresented in the policymaking process in the American states. [Table 3 about here] As discussed above, the primary rationale behind examining unequal political representation at the state level is to understand and explain variation in political equality across the states. To assess in which states political influence is strongly tied to income compared to those states that weight opinions more equally, I run a separate regression for each state and 13 Because residents are clustered within states and experience the same state policy, I report standard errors clustered by state for all regressions in Table 3. The results are substantively similar if a multilevel model (with respondents nested within states) is used instead. 16

19 compare the coefficient for (state relative) income. Similar to the nationwide regression reported above, a more steeply negative slope coefficient indicates a stronger relationship between income and ideological distance and, accordingly, less political equality. For example, consider the two hypothetical states presented in Figure 2. For each state, the line represents the slope of the relationship between income and ideological distance. As the figure illustrates, the relationship between income and distance is rather weak in State C, indicating that citizens opinions are weighted roughly equally regardless of their income. In contrast, the slope of the relationship between income and ideological distance is quite steeply negative for State D, indicating that there is a strong degree of political inequality in state policymaking. 14 [Figure 2 about here] I run a separate regression for each state using each of the six different measures of ideological proximity described above (three measurement techniques x two measures of state policy liberalism). 15 When the six regression coefficients (for state relative income) are 14 This approach of running separate regressions for each state and comparing the slope coefficients is methodologically similar to Gelman, Park, Shor, Bafumi, and Cortina s (2008) analysis of the varying relationship between income and partisan vote choice across the states. 15 One potential concern with running a regression separately for each state with opinion-policy distance as the dependent variable is that every respondent has the same value for state policy, effectively making the policy term a constant. However, consider a state where income and ideological conservatism correlate perfectly (i.e. as income increases, so does ideological conservatism). If the state s policy position is more conservative than all citizens ideology positions, the regression coefficient for income would be negative (indicating that as income increases, ideological distance between opinion and policy decreases). But, if the state s policy position is more liberal than all citizens ideology positions, the coefficient for income would be positive (indicating that as income increases, ideological distance 17

20 compared within the states, they have a Cronbach's alpha of.96, indicating that all six measures appear to be measuring the same underlying concept. To create a single summary score of political equality that is directly comparable across states, I conduct a principal components analysis on the six slope coefficients and generate a single factor score for each state. 16 Because a more steeply negative slope coefficient indicates more unequal representation, a more positive factor score indicates greater political equality. I label this new measure the Political Equality Index. [Table 4 about here] The factor scores generated using this procedure are reported in Table 4 where the states are ranked from the most to least equal in terms of political representation. It is important to note that the index is not simply an alternative measure of the liberalism of state policy (with the expectation that lower income citizens support more liberal policies). The Political Equality Index correlates with the Gray et al. (2004) policy liberalism measure at.47 and with the Sorens, Muedini, and Ruger (2008) policy liberalism measure at.only 37. Most importantly, however, is the fact that there is significant variation in political equality across the states. In the following section, I use this variation to evaluate whether states that allow for direct democracy tend to more equally weigh citizens opinions in the state policymaking process. between opinion and policy also increases). Even though the distribution of citizens opinions is identical under both scenarios, the regression coefficients are very different depending on where state policy is located in the ideological space (relative to citizens opinions). Therefore, the coefficient for respondents income for single state regressions does not simply indicate the relationship between income and ideology within a state but instead indicates (as intended) the sign and strength of the relationship between income and opinion-policy distance. 16 The eigenvalue for the lone retained factor is 5.15 and explains 86% of the total variance. 18

21 Does the Ballot Initiative Promote Political Equality? To evaluate the relationship between direct democracy and political equality, I examine several different measures of initiative use. To begin, I generate a simple dummy variable that is coded one if a state has the initiative process and zero if it does not (Boehmke 2002 and Matsusaka 2004 also uses this technique). Fully 24 of the 50 states have the process (see Table 1), but only 23 of the states in the models I subsequently present do because Alaska drops out due to no public opinion data (and hence no measure of the equality of political representation). I regress the Political Equality Index (where greater equality is coded higher) on this dummy variable for presence of the initiative process. I also account for two economic variables that are likely related to political inequality. First, I include a measure of a state s median income (measured in $1000s) with the expectation that richer and poorer states may differ in the way economic inequality is reproduced in politics and in terms of the issue cleavages that define political conflict for both voters and elites. This expectation builds on Rigby and Wright s (2011) finding that political representation is most unequal on economic issues in poorer states (where it is the main axis of political conflict), suggesting that the degree of political inequality is conditioned by a state s wealth Moreover, Gelman et al. (2008) persuasively demonstrate that the relationship between citizens incomes and their partisan voting behavior varies widely depending on the aggregate wealth of one s state of residence. Specifically, they find the strongest relationship between income and vote choice (such that the probability of voting Republican increases as one s income increases) in poorer states like Mississippi while the relationship is attenuated almost to zero in richer states like Connecticut. 19

22 Second, I include a measure of income inequality for each state using the Gini coefficient for 1999 provided by the United States Census Bureau. The Gini coefficient is a commonly used measure of income inequality that runs from zero to one, with higher values indicating the income distribution is more unequal and concentrated in the hands of a small group of citizens. Bartels (2008) recent work on the political economy of the new gilded age posits a close relationship between political and economic inequality. Moreover, in a cross-national analysis, Giger, Rosset, and Bernauer (2012) find that political parties are especially unresponsive to the opinions of poor citizens in countries with higher levels of economic inequality. The results from this first OLS regression estimation are reported in Column 1 of Table 5. The coefficient for the presence of the initiative process is positive, but not statistically different from zero. This suggests that states with the initiative process do not systematically differ in their level of political equality compared to states that do not have the initiative process. Looking to the other covariates in the model, I also find that, as expected, states with lower levels of income inequality tend to weigh citizens opinions more equally than states with wider income differences. [Table 5 about here] The splitting of states into those with the initiative process and those without does not tell the whole story about the relationship between direct democracy and political inequality, for several reasons. First, states that have the ballot initiative process may qualitatively differ in several aspects (many of them unobservable) from states that do not (Keele 2009). 18 Second, 18 For example, Western states are far more likely to have the ballot initiative process than states in other regions of the United States. So, analyses that uncover differences between initiative and non-initiative states may instead be picking up on regional differences in political culture/history. 20

23 among states with the process, some may routinely have multiple policy proposals on the ballot for voters to decide on while others may rarely, if ever, use the initiative process. Third, the very laws that decide how a proposal makes it onto the ballot vary considerably across the states. As Bowler and Donovan (2004, 348) point out, Although the initiative process follows a very similar sequence across the states (titling, qualification through petition, vote), there are critical differences among the states rules for its implementation that structure the cost and difficulty of qualifying a measure for the ballot. These differences provide a more nuanced picture of the initiative process and allow me to examine whether, across only the states that have the initiative process, legal processes and frequency of use predicts the extent of unequal political representation. To quantify this variation, I use two variables developed by Bowler and Donovan (2004) that measure the institutional constraints on the initiative process across the states. The first is a Qualification Difficulty Index, measured as the sum of the number of formal provisions that increase the difficulty of qualifying a measure for the ballot, giving special weight to a state s petition signature requirements (320). The second is a Legislative Insulation Index, measured as the sum of the number of provisions that constrain how a legislature can change an initiative that has been approved by voters (320). The variables are coded such that a higher value indicates that access to the ballot is more difficult or that the state legislature has a greater ability to modify the content of successfully petitioned ballot proposals. The Political Equality Index is regressed (separately) on each of these two institutional indices only among the 23 states that have a legal initiative process. The results of these estimations are reported in Columns 2 (Qualification Difficulty Index) and 3 (Legislative Insulation Index) of Table 5. The coefficients for both institutional indices are negative, but only 21

24 the coefficient for Qualification Difficulty Index is bounded below zero. This indicates that states where it is more difficult to actually place an initiative on the ballot tend to weigh citizens opinions less equally. The magnitude of this relationship is substantively large: moving from one standard deviation below the mean to one standard above on the Qualification Difficulty Index leads to a.78 standard deviation decrease in the Political Equality Index. In short, these findings suggest that states that limit the impact of the initiative process are also more politically unequal. As another way to evaluate the relationship between the ballot initiative and political equality, I construct two measures of actual initiative usage using data from the Initiative and Referendum Institute at the University of Southern California: (1) the number of ballot initiatives proposed (i.e. placed on the ballot for voters to decide on) from 1990 to 2008 and (2) the number of ballot initiatives actually approved by voters during that same time period. 19 These measures that encompass an eighteen year period are intended to a give general indication of how often (or not) the initiative process is used in a particular state. The Political Equality Index is regressed (separately) on the number of ballot initiatives proposed and approved and the same controls for state median income and level of income inequality used in the previous models. 19 The number of initiatives proposed ranges from 0 (Illinois) to 124 (California) with a mean of and a standard deviation of The number of initiatives approved ranges from 0 (Illinois and Mississippi) to 44 (California) with a mean of and a standard deviation of The correlation between these two usage measures (the number of initiatives proposed and approved) and the two institutional measures (the Qualification Difficulty Index and the Legislative Insulation Index) never exceeds -.60 (the correlations are negative because more restrictions are related to less usage), which indicates that the measures are related but not identical. 22

25 The results of these estimations are reported in Columns 4 (number of initiatives proposed) and 5 (number of initiatives approved) of Table 5. Both coefficients for actual initiative usage are positive and bounded above zero. This suggests that states that use the ballot initiative process more often tend to be more politically equal, and the magnitude of this relationship is substantively large. Specifically, moving from one standard deviation below the mean to one standard deviation above for the number of ballot initiatives proposed predicts a.88 standard deviation increase in the Political Equality Index. Similarly, moving from one standard deviation below the mean to one standard deviation above for the number of ballot initiatives approved predicts over a full standard deviation (1.08) increase in the Political Equality Index. In sum, when the analysis is confined only to states with the initiative process and actual usage is considered, states that make more frequent use of the ballot initiative process tend to have more equal political representation. 20 Discussion The correspondence between citizens opinions and public policy is the bottom line for American democracy. A large political science literature has been dedicated to demonstrating that citizens aggregated opinions strongly predict the tone of public policy in both state (e.g., 20 The models reported in Columns 4 and 5 of Table 5 include Illinois as a state that has the initiative process even though no initiatives were proposed or approved in that state from 1990 to To make sure that inclusion of Illinois is not driving the results, the same models were estimated without Illinois included in the analysis. The results are substantively identical. I also regressed (separately) the Political Equality Index on the logged values of number of ballot initiatives proposed/approved and again found substantively identical results. 23

26 Erikson, Wright, and McIver 1993; Lax and Phillips 2012) and national politics (e.g., Erikson, MacKuen, and Stimson 2002). Far less attention has been paid to the question: Are citizens opinions represented equally? Recent studies at the national level (Gilens 2005, 2012; Jacobs and Page 2005; Bartels 2008; Ellis 2012) report that the opinions of the disadvantaged are especially underrepresented in the policymaking process compared to the affluent across a wide array of policy domains. This paper extends this new line of inquiry to the American states and uncovers similar results (also see Gilens, Lax, and Phillips 2011; Rigby and Wright 2011; Flavin 2012). Assessing the relationship between citizens general political ideology and state policy liberalism, citizens with higher incomes are consistently better represented compared to citizens with lower incomes (see Table 3). If a key characteristic of a democracy is the continued responsiveness of the government to the preferences of its citizens, considered as political equals (Dahl 1971, 1), the democratic process in the American states appears to fall short of this standard. Moreover, there is significant variation in the equality of political representation across the states (see Table 4). Using this variation, I evaluate the relationship between direct democracy and political equality and first find that states with the ballot initiative process are no more or less politically equal compared to states without the initiative process. However, when the analysis is confined only to states with the initiative, public policy outputs are more equally representative of all citizens opinions in states where it is easier to place a measure on the ballot for popular vote and states where the initiative process is heavily used (see Table 5). These results suggest that having and frequently using the initiative process may be a viable avenue for ensuring that the opinions of disadvantaged citizens are represented in the political arena. 24